Landscape complexity and vegetation dynamics in Riding Mountain National Park, Canada
Walker, David John
The primary focus of landscape ecology is the interrelationship between spatial pattern and processes within an ecosystem. It is through their mutual interaction that landscape structure and complexity are ultimately determined. Complexity, which includes both the horizontal and vertical arrangement of vegetation structure on the landscape, is an emergent property of dynamic systems. In the boreal forest, landscape complexity is a product of successional dynamics, physiography and environmental variability. The objective of this study was to examine spatial and temporal changes to landscape complexity in the boreal mixedwood of Riding Mountain National Park (RMNP), Canada. Using remotely sensed Landsat data and scale invariant fractal measures of spatial pattern, change in landscape complexity under natural and human induced fragmentation regimes was examined. The importance of structure as an emergent property of boreal canopies and its influence on landscape mapping using satellite data was addressed. It was found that landscape-level spatial pattern became increasingly entropic during succession. Old landscapes (120 years post-fire) were typified by a landscape matrix dominated by small scale patches and low spatial persistence. Physiography was also found to influence scale invariant landscape complexity. Landscapes typified by simple physiographies (well-drained, topographically simple sites) were characterized by a few dominant over-dispersed land-cover classes. Complex landscapes (variably drained, topographically complex sites), patches were under-dispersed and contagious, however complex gradients resulted in high pattern complexity (increased juxapositioning of landscape elements). It is suggested that the accumulation of small-scale disturbances over time and species turnover along complex environmental gradient affect high landscape complexity in the boreal forest. In contrast, human driven disturbance processes in the boreal forest resulted in lower spatial complexity over time. Fragmentation and habitat losses in the region surrounding RMNP were found to be high, with only half of the forest present in 1950 remaining in the 1990's. Scale-invariant spatial dispersion of forest fragments decreased between the 1950's and 1990's. Thus, the study area is becoming increasingly isolated from other natural forested areas within the region. In creating maps of land cover for these analyses, it was found that structural composition of the canopy was often more important than floristics in determining spectral reflectance in Landsat data. A rule-based optimization procedure using multivariate analysis was developed to maximize the relationship between vegetation on the ground and spectral reflectance. Because of the high degree of spatial complexity in these systems, an alternative approach to map accuracy assessment utilizing multiple discriminant analysis (MDA) was developed. It was found that closed conifer stands composed of different softwood species were not easily discriminated during classification because of identical spectral signatures at the stand-level. It is suggested that the highly structured architecture and conical form of conifer stands results in the anechoic interception and absorption of light. This light interception strategy may have adaptive advantages in regions where sun angle is low, or where cloud cover is high, such as in the boreal forest and montane environments. The results of these investigations into landscape pattern suggest that ecosystem dynamics in the boreal forest produce scale-invariant landscape complexity.